Newsgroup sci.stat.consult 21792

Directory

Can someone help me with some SAS commands please?
I have a 2x4 mixed repeated ANOVA. I have a between subjects manipulation, ITK,
(2 levels) and each subject goes through 4 treatments (and so I have 4
scores from each subject). I found a significant interactive effect.
I want to do
(1) contrast tests, and
(2) trend tests for linear, quadratic and  cubic.
I looked through the SAS manuals but I could not figure out the way to do
these tests.
The following is part of the SAS program I have been using:
DATA ONE;
INFILE 'INTRA DATA A';
/* A DESCRIPTION OF THE VARIABLES IN INTRA*/
INPUT ITK SUBJNO Y1 Y2 Y3 Y4;
PROC PRINT;
PROC GLM DATA=ONE;
CLASS ITK;
MODEL Y1-Y4 = ITK;
REPEATED TIME 4 / SUMMARY;
MEANS ITK;
RUN;
TIA.
-- Kim Tan (PhD student, Temple University).

Whether one uses a one-way or two-way anova to calculate ICC depends on
the theoretical status of any differences in means between raters.  In the
one-way approach, the single factor is always the sampling units (e.g.,
subjects, samples) that were measured or rated.  The error in this case is
made up of any differences in the means of raters as well as the
interaction of subjects X raters.  If there are good reasons to exclude
variation in rater means from the error term, then a two-way anova, where
the factors are subject and rater, would be appropriate.  The latter
analysis removes systematic biases between raters from the error term, and
leaves only the rater X subject interaction.
An article that I often refer my clients to is Tinsley, H. E. A., & Weiss,
D. J. (1975). Interrater reliability and agreement of subjective
judgments. Journal of Counseling Psychology, 22, 358-376.  This gives a
more thorough discussion of this topic.
Christina M. Gullion
In article
,
Chauncey Parker  writes:
>Subject:	ICC:interrater reliability; ?anova models?
>From:	Chauncey Parker 
>Date:	Sun, 22 Dec 1996 11:31:40 -0800
>
>I see articles talk about using onway vs two way anova error terms to
>calculate ICC. However, It seems that all cases use the error terms from
a
>"single factor repeated measure anova" that provides error terms for BMS
&
>WMS and the WMS is partitioned into JMS and EMS.
>
>first, am I right to see this anova (the sing fact repeated meas) as a
>oneway? 
>
>and, this is synonomous with a within subs oneway anova?
>
>
>OK, what is the deal with the talk about two way anovas and when, if
ever,
>do you use:
>
>two way anova?
>
>
>simple oneway anova (gives only BMS & WMS terms)?
>
>
>Thanks a clinically significant amount . . .    =;/
>Chauncey@UW

As far as I can make out. you appear to be doing an analysis of covariance
(ancova) with needle density as the covariate, family as the class
variable, and egg counts as the dependent variable.
Whether needle density is a continuous variable or not may not be the
right question for your analytic problem.  Even though you have chosen to
"put" your needle density data into 6 bins, density probably is
continuous.  Of more relevance is whether your data are approximately
symmetrically distributed over the 6 points, or are sharply skewed to left
or right.  Differences between families in the variability of the needle
density ratings is likely to be your biggest problem in analyzing and
interpreting your experimental results.  Also, the method you are using
assumes that the regression slopes of needle density on egg count are
equal across the families.  This is unlikely to be the case if some
families have sharply different needle density distributions.
What you are doing is not a simple analysis.  You would be wise to find a
statistician who could lead you through the tests of assumptions and
diagnostics that are needed to do an ancova properly.
Best wishes,
Christina M. Gullion, PhD

Whether you can compare the two surveys depends on the substantive
comparison between the 5-point and 7-point scales.  If both refer to the
same theoretically CONTINUOUS underlying concept or construct but simply
have a different number of anchors, then it may be possible to standardize
the longer scale to 5 points with a simple linear transformation.  
For example, if in both years you had a scale running from strongly
disagree to strongly agree, but you put a couple more intermediate points,
then it may be reasonable to rescale.  An even simpler case is if the
measures were done essentially on "rulers", e.g.,  -2 -1 0 1 2  vs -3 -2
-1 0 1 2 3.  Multiplying ratings on the latter scale by 2/3 will put them
on the same -2 to +2 scale as the shorter ruler.
This approach assumes that the longer scale does not "reach" further into
the extremes, but simply partitions the continuous distribution into more
bins.
If, however, the two surveys represent different conceptualizations of the
dimension being measured or if the scales had specific anchor points,
which differ between years, then you cannot compare the two surveys--they
were measuring different things.  
In article <5ah83k$sb6@boursy.news.erols.com>, Claude Lijoi
 writes:
>Subject:	scale comparison
>From:	Claude Lijoi 
>Date:	2 Jan 1997 21:08:04 GMT
>
>I want to track results of a survey. Problem is that last year's survey 
>was on a five point scale and this year's is on a seven point scale. What
>is the best way to convert from one form to another for comparison? I am 
>an SPSS user.
>
>Thanks for the interest,
>
>Claude Lijoi
>cglijoi@erols.com